GLAST LAT DC1 closeout February 12-13, 2004 DC1 lesson learned Francesco Longo University and INFN, Trieste, Italy francesco.longo@ts.infn.it F.Longo DC1 lessons learned 1 GLAST LAT DC1 closeout February 12-13, 2004 Outline • • • • • • • DC1 purposes and scope DC1 preparation LAT data analysis needs – Background rejection and OnBoardFilter – Gamma-ray Analysis DC1 data Science Tools for DC1 – Architecture – Distribution – Documentation DC1 “community” – The Wiki page – DC1 mailing list What’s next? – Analysis of DC1 data F.Longo DC1 lessons learned 2 GLAST LAT DC1 closeout February 12-13, 2004 Purposes of the Data Challenges S.Ritz • “End-to-end” testing of analysis software. • Familiarize team with data content, formats, tools and realistic details of analysis issues (both instrumental and astrophysical). • If needed, develop additional methods for analyzing LAT data, encouraging alternatives that fit within the existing framework. • Provide feedback to the SAS group on what works and what is missing from the data formats and tools. • Uncover systematic effects in reconstruction and analysis. Support readiness by launch time to do all first-year science. F.Longo DC1 lessons learned 3 GLAST LAT DC1 closeout February 12-13, 2004 Lessons from Data Challenge 1 • • • • • “End-to-end” testing of analysis software. – First usage of “key tools” (Likelihood, GRB tools) Familiarize team with data content, formats, tools and realistic details of analysis issues (both instrumental and astrophysical). – FITS, FTOOLS stuff, Exposure, TS maps, Binning, Spectral Analysis, Count rate triggers … – Background rejection and StdCut events – Galactic Plane Modeling If needed, develop additional methods for analyzing LAT data, encouraging alternatives that fit within the existing framework. – Tools development Provide feedback to the SAS group on what works and what is missing from the data formats and tools. – User feedback Uncover systematic effects in reconstruction and analysis. – DC1 preparation First attempt to do LAT science for many users F.Longo DC1 lessons learned 4 GLAST LAT DC1 closeout February 12-13, 2004 Data Challenge Planning Approach S.Ritz • Walk before running: design a progression of studies. • DC1. Modest goals. Contains most essential features of a data challenge. – 1 simulated day all-sky survey simulation • find GRB • a few physics surprises • exercise: – exposure, orbit/attitude handling, data processing pipeline components, analysis tools • DC2, start end of CY04. More ambitious goals. Encourage further development, based on lessons from DC1. One simulated month. • DC3. Support for flight science production. F.Longo DC1 lessons learned 5 GLAST LAT DC1 closeout February 12-13, 2004 What’s been done: preparatory work S.Ritz • Very large effort during the past ~9 months by many people. • Instrument analysis: – done previously with earlier tools for AO, PDR, etc., demonstrating LAT meets requirements. – Now done again with new tools. More to do, but more than adequate for DC1. • Fluxes • Data formats, processing • Science tools Already a great success! F.Longo DC1 lessons learned 6 GLAST LAT DC1 closeout February 12-13, 2004 DC1 R.Dubois • Focal point for many threads – Orbit, rocking, celestial coordinates, pointing history – Plausible model of the sky – Background rejection and event selection – Instrument Response Functions – Data formats for input to high level tools – First look at major science tools – Likelihood, Observation Simulator – Generation of datasets – Populate and exercise data server at SSC & LAT – Code distribution on windows and linux • Involve new users • Teamwork! F.Longo DC1 lessons learned 7 GLAST LAT DC1 closeout February 12-13, 2004 Backgrounds • • • S.Ritz DC1 is an approximate modeling of one day of LAT data. – at face value, this means 400M background triggers (4kHz). [This is not the number generated, which is larger, since many miss the instrument.] – using an updated version of Bill Atwood’s background rejection analysis shown in Rome, residual contamination of photon sample would be about ~6%. One-day science is generally NOT background limited. Several purposes to generating background for DC1: – exercise the machine, find the problems (already done!) – generate amounts of background needed anyway to complete the analysis. We therefore decided to unhook the background generation for DC1 – Rejection analysis already at a sufficient level to estimate instrument performance for gammas (Aeff). – Used these cuts on the photon sample for DC1. Provides a good description of impacts of background rejection. • at normal incidence, Aeff asymptotes to 10,000 cm2. At 100 MeV, ~4,500 cm2 • small fall-off in area for E>10 GeV, will be improved soon. Not a background rejection issue. • – Already at a sufficient level to make background a non-issue for DC1 science analyses. – This allows the background rejection analysis to proceed at its own pace. At end of DC1, both background rejection and signal analyses will be completed to the levels planned. F.Longo DC1 lessons learned 8 GLAST LAT DC1 closeout February 12-13, 2004 Std Cut files • Why from dataServers not the stdCut file as default and the other data only as option? • Use as default the events that passed the background, PSF, and energy resolution filters (i.e., the events for which the DC1 response functions apply) • More kinds of response functions (e.g., a set that applies to the events that don't pass the PSF filter), but even so the event flags should be accessible in the data servers. • Bias on Analysis! F.Longo DC1 lessons learned 9 GLAST LAT DC1 closeout February 12-13, 2004 DC1 lesson (already) learned • • • • • Cf S.Ritz Lots of hard work by many people on the machinery up until the last minute. – Richard’s talk Sky model is fairly rich and accurate. Some details could be better – no flaring sources implemented (though some variable sources might have day-long fluxes different from your expectation…) – 2 GRB “models” available No (intentional) hardware problems implemented. – decided to postpone to DC2, when ISOC is up and running No onboard filter in data path yet – similar to background rejection situation. The incremental loss of area (after other cuts) is now expected to be very small. Instrument response functions are not really mature. – some problems still. certainly good enough for DC1 science goals, however! F.Longo DC1 lessons learned 10 GLAST LAT DC1 closeout February 12-13, 2004 DC1 data • FITS versus ROOT data • Spacecraft pointing info only in FITS? • Pointing history available as .dat or as separate tree in ROOT file? • ROOT data server? F.Longo DC1 lessons learned 11 GLAST LAT DC1 closeout February 12-13, 2004 DC1 Minimum Results • • S.Ritz The existence of the data sets and the volume of data generated for background analyses already meets one of the success criteria. A minimum set of plots and tables that we must collectively produce: – TABLE 1: found sources, ranked by flux (E>100 MeV). Table has the following columns – – – – – reconstructed location and error circle flux (E>100 MeV) and error significance 3EG identification (yes or no) [note: DON’T assume DC1 sky is the 3EG catalog!] extra credit: » include flux below 100 MeV » spectral indices of brightest sources » comparison of 3EG position and flux characteristics with GLAST analysis – FIGURE 1: LogN-logs plot of TABLE1 – TABLE 2: list of transients detected. Columns are • location and error circle • flux (E>100 MeV) and error • significance • duration – FIGURE 2: light curve • Extra credit: FIGURE 2a: spectra. – PLUS: reports of any physics surprises found. F.Longo DC1 lessons learned 12 GLAST LAT DC1 closeout February 12-13, 2004 MC Truth unveiled • Scripting for Catalog generation • How to analyze new sources? E.g. the Galactic Center • Connection among different tools e.g. blind searching tool and likelihood analysis • Take into account instrument behaviour in different ranges • Success of GRB trigger algorithms F.Longo DC1 lessons learned 13 GLAST LAT DC1 closeout February 12-13, 2004 Source Detection • • • • • Many methods developed (wavelets, voronoi tesselation..) Need compare the results Likelihood result on blind search detected sources Generation of catalog with different methods significance? Study of Interstellar emission F.Longo DC1 lessons learned 14 GLAST LAT DC1 closeout February 12-13, 2004 GRB analysis • • • • • • • • • • Lot of work in Italy and US 5 different trigger criteria Need a comparison Need to study onboard LAT trigger performances (e.g. for on ground follow up) Need to study weaker GRB near the detection limit and optimize trigger with realistic background Dividing the sky in several spatial bin is an effective way to reduce background Floating threshold and trigger window selections (time, events, …) to be refined with variable background rates On board buffer with localization and timing info available for unbinned search on board Effects of CR on localization accuracy could be solved by weighting the events by their distance to accumulated centroid Need refinements in Spectral analysis F.Longo DC1 lessons learned 15 GLAST LAT DC1 closeout February 12-13, 2004 Beyond the minimum S.Ritz • Here are a few suggestions: – we may generate and release more days of data • better exercise tools and infrastructure • more transients – spectral analyses – localization studies • one-day localization of Vela is particularly interesting – analysis improvements • But don’t let this list limit you. The sky is the limit! F.Longo DC1 lessons learned 16 GLAST LAT DC1 closeout February 12-13, 2004 Beyond minimum results • Source detection methods • New analysis methods and/or languages • Little usage of Science Tools? F.Longo DC1 lessons learned 17 GLAST LAT DC1 closeout February 12-13, 2004 High-Level Analysis S.Digel and P.Nolan Science Tools (Apologies to R. Dubois, D. Flath, M. Urry, P. Padovani) (T. Usher) Seth Digel & Patrick Nolan HEPL/Stanford Univ. F.Longo DC1 lessons learned 18 GLAST LAT DC1 closeout February 12-13, 2004 DC1 Analysis Science Tools Other Tools(?) (Apologies to R. Dubois, D. Flath, R.Roy Britt) (T. Usher) F.Longo DC1 lessons learned 19 GLAST LAT DC1 closeout February 12-13, 2004 DC1 Tools S.Digel and P.Nolan • Data Challenge 1 is not about science – The emphasis is on the analysis techniques themselves and the tools to implement them • What are the DC1 goals for the science tools? – Briefly, an end-to-end test of the system, going back to instrument simulation, event classification and generation of response functions, through high-level processing, and managing the flow of data in between – So we are at the high-level processing end of the chain now (more or less – will be some iteration, e.g., on event classes). We want to • shake test the science tools; actually they won’t take a lot right now • introduce the analysis methods for gamma-ray astronomy with the LAT • get feedback on functionality – from users and from developers F.Longo DC1 lessons learned 20 GLAST LAT DC1 closeout February 12-13, 2004 S.Digel and P.Nolan Science Tools in DC1 DC3 The big picture: Details are changing, but still basically right Level 0.5 Level 1 (D1) Pointing/livetime history (D2) Alternative source for testing highlevel analysis Event display (UI1) Pulsar ephem. (D4) Pulsar period search (A4) Ephemeris extract (U11) Arrival time correction (U10) Pulsar phase assign (U12) Pulsar profiles (A3)1 Data extract (U1) Pt.ing/livetime extractor (U3) Alternative for making additional cuts on alreadyretrieved event data Observation simulator (O2) Data subselection (U2) Pt.ing/livetime simulator (O1) Pt.ing/livetime extractor (U3) Exposure calc. (U4) This tool also performs periodicity tests and the results can be used to refine ephemerides 2 These tools can also take as input binned data from other instruments, e.g., GBM; the corresponding DRMs must also be available. LAT Point source catalog (D5) Source model def. tool (U7) Catalog Access (U9) Likelihood (A1) Astron. catalogs (D6) Src. ID (A2) Interstellar em. model (U5) Map gen (U6) IRFs (D3) IRF visualization (U8) GRB LAT DRM gen. (U14) 1 F.Longo Standard Analysis Environment GRB event binning (A5) GRB spectral analysis (A8)2 GRB rebinning (A6)2 DC1 lessons learned User Interface aspects of the standard analysis environment, such as GRB unbinned Image/plot display (UI2), Command spectral analysis (A9) line interface & scripting (UI4), and GUI & Web access (UI5) are not GRB spectral-temporal shown explicitly. modeling (A10) GRB visualization (U13) GRB temporal analysis (A7)2 14 Sept 2002 21 GLAST LAT DC1 closeout February 12-13, 2004 S.Digel and P.Nolan • Science Tools in DC1 All components are still prototypes Level 1 (D1) Pointing/livetime history (D2) Alternative source for testing highlevel analysis Data extract (U1) Pt.ing/livetime extractor (U3) Alternative for making additional cuts on alreadyretrieved event data Observation simulator (O2) Data subselection (U2) Pt.ing/livetime simulator (O1) Pt.ing/livetime extractor (U3) Source model def. tool (U7) Exposure calc. (U4) Likelihood (A1) Interstellar em. model (U5) Map gen (U6) IRFs (D3) GRB LAT DRM gen. (U14) The DC1 functionality is Data extraction Limited visualization Model definition Model fitting Observation simulation GRB spectral analysis (A8)2 GRB event binning (A5) F.Longo DC1 lessons learned 22 GLAST LAT DC1 closeout February 12-13, 2004 Implementation of Science Tools S.Digel and P.Nolan • Reminder: the tools are implemented as FTOOLS – HEASARC convention across missions • You will notice that this defines much of the ‘look and feel’ – Provides a uniform interface – HOOPS for prompting at the command line – GUI is coming • FITS files for data F.Longo DC1 lessons learned 23 GLAST LAT DC1 closeout February 12-13, 2004 Prototype Science Tools in DC1 • Many tools planned to be included in SAE not yet available • GRB Tools – Trigger methods – GRB visualisation – GRB spectral analysis – GRB temporal analysis • Exposure and Map generation • Alternative source detection methods • ROOT tools F.Longo DC1 lessons learned 24 GLAST LAT DC1 closeout February 12-13, 2004 DC1 communication issues • Usage of Wiki page – a “cascade” model • DC1 mailing list – ScienceTools usage mailing lists? • cvs repository for DC1 tools? – Use the /users/ tree F.Longo DC1 lessons learned Wiki usage (Riccardo) 25 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • • • • • • • • Astrophysical data analysis Software usage and reliability Documentation Data access and data server usage UI stuff Software installation and release Software infrastructure & framework Communication and Time frame F.Longo DC1 lessons learned 26 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Analysis issues – First analysis on Astro data – Transients: • Notification for deep observation • Not requested to be run by every user • Need of a Quick Look tool for DC2 study – Likelihood tool: • How to check fit quality? • Which statistical minimum for reasonable results? • Help from experienced users – In some cases incorrect FITS headers – Definition of Galactic diffuse emission (model, units …) – SLAC installation of common Astro tools (FTOOLS, ds9, XSPEC..) F.Longo DC1 lessons learned 27 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Software: – Clear documentation for most tools – More files greater the possibility to do mistakes (could be FT2 information embedded in FT1?) – FT1 data and Exposure in different directions tests? – Need for Sanity checking on the inputs. (e.g. Likelihood tool with 'BACK' response functions without event filtering) – TsMap crashes – Different optimizers different results? – Most self-contained not help for developing own tools that fit within framework (e.g. read/visualize data, manipulate multidimensional arrays, get background model density at a given point, access the IRFs etc etc). Need more documentation on existing “base” tools – How to generate different Energy Bin fits file? – Plot tools within the SAE and exposure maps generation missing – ROOT format for exposure map or more general exposure tool – Coordinates specification (e.g. for modeldef) – CPU time F.Longo DC1 lessons learned 28 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Software reliability – Statistical interpretation of Likelihood – Need for comparison among different analysis F.Longo DC1 lessons learned 29 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Documentation: – Wiki pages good idea – Fine documentation of the science tools (for Linux users?) – Not updated to new versions – Need for dynamic documentation (e.g. galactic diffuse for obsSim) – Weekly status report? – Needs for tips on developing own models – Info on tool contact and updates on each web page F.Longo DC1 lessons learned 30 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Access to Data: – Some failures – Documentation on SSC better than SLAC data server (e.g. time selection units?) – Cut on tree variables directly? – Full data set available not from DataServers F.Longo DC1 lessons learned 31 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Coherent User Interface – Who used the Likelihood GUI? – At least command line with arguments from arbitrary directory – Unique settings startup command for analysis session – Simple usage explanation (e.g. “– help”) – Tab completion or command history available – User should not worry about data and executable location – High priority for DC2 for ST to be distributed to the general community. F.Longo DC1 lessons learned 32 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Software Installation – Binary distributions (v1r0p1) very nice but which Linux version? (7.3 vs 9.0?) – New releases not publicly distributed as binary… Only “core” developers – Necessity for knowing how to include new features in newer version of packages – Windows vs Linux compilation and distribution F.Longo DC1 lessons learned 33 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Software infrastructure – Bookkeeping Info (events N, area, Energy spectrum parameters, energy range for events) – Messages from ScienceTools s/w (Info, Warning, etc ) • Level control, name of algorithm, service, .. • Std message format – Setup of environment by automated way .(c)sh scripts not easily maintainable – PFILES definition for unique parameter file directory – Coding rules for avoiding compilation warnings F.Longo DC1 lessons learned 34 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • Framework – FTOOLS structure for distributed software – How to extend or use other tools? – How to include new tools? F.Longo DC1 lessons learned 35 GLAST LAT DC1 closeout February 12-13, 2004 DC1 feedback • General Things: – Communication issues: (FT1Energy fix, updates to data servers, new ST releases?) – Before holiday work caused net loss of time F.Longo DC1 lessons learned 36 GLAST LAT DC1 closeout February 12-13, 2004 Proposal • Science Tools user meetings? • Continue to analyse DC1 data before DC2! Voronoi tesselation of GRB Example of usefulness of communication F.Longo DC1 lessons learned 37 GLAST LAT DC1 closeout February 12-13, 2004 Conclusions • • • • • • DC1 a success! First analysis of astrophysical data for many of us Science Tools developments Work on Background and Filter DC1 data generation and storage Team collaboration F.Longo DC1 lessons learned 38 GLAST LAT DC1 closeout February 12-13, 2004 Acknowledgements • Thanks to those who sent me comments • Steve, Julie, Dirk, Toby, Jay, Jerry, Traudl, Seth, Benoit, Nicola, Gino, Claudia, Riccardo, Alessandro, Michael, Francesca, Monica, Luca F.Longo DC1 lessons learned 39 GLAST LAT DC1 closeout February 12-13, 2004 Science Tools in DC1 S.Digel and P.Nolan • Details of the contents of the event summaries are still converging • The interstellar emission model is still the model used by EGRET team (Bertsch et al. 1993, Hunter et al. 1997) • The instrument response functions are defined only for the events that pass the filters presented by Bill Atwood at the collaboration meeting in Rome – Eventually we expect to have more than one event class – PSF and energy resolution are being defined on a grid of energies and inclination angles, with a analytic function fit. No interpolation of the parameters between grid points is attempted – The IRFs are not yet in CALDB, although this switch should be completely transparent to the user • The interface to EGRET data and pointing/livetime history is not complete (and was not planned for DC1) F.Longo DC1 lessons learned 40 GLAST LAT DC1 closeout February 12-13, 2004 Science Tools in DC1 S.Digel and P.Nolan • The DRM generator RspGen understands only circular cutout regions so far; this is not a limitation at all for analyzing bright GRBs, but the intent is to make it understand custom shapes for crowded fields • The map generation tool does not exist yet – EventBin (GRB event binning) can make counts maps – Exposure maps can be generated with [what] • No visualization is integrated with the tools yet • The orbit and attitude simulation is still idealized and not yet a standalone tool F.Longo DC1 lessons learned 41